5.6-5.10 balance regression?
Hello,
As part of the maintenance routine of one of my raid1 FS, a few days ago I was in the process of replacing a 10T drive with a 16T one.
So I first added the new 16T drive to the FS (btrfs dev add), then started a btrfs dev del.
After a few days of balancing the block groups out of the old 10T drive, the balance aborted when around 500 GiB of data was still to be moved out of the drive:
Dec 21 14:18:40 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:18:54 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:19:16 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2
Of course this also cancelled the device deletion, so after that the device was still part of the FS. I then tried to do a balance manually, in an attempt to reproduce the issue:
Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): balance: start -ddevid=5,limit=1 Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:28:29 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:28:46 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2
There were of course still plenty of room on the FS, as I added a new 16T drive (a btrfs fi usage is further down this email), so it struck me as odd.
So, I tried to lower the reduncancy temporarily, expecting the balance of this block group to complete immediately given that there were already a copy of this data present on another drive:
Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1 Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:39:00 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:39:17 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2
That didn't work.
I also tried to mount the FS in degraded mode, with the drive I wanted to remove missing, using btrfs dev del missing, but the balance still failed with the same error on the same block group.
So, as I was running 5.10.1 just for a few days, I tried an older kernel: 5.6.17, and retried the balance once again (with still the drive voluntarily missing):
[ 413.188812] BTRFS info (device dm-10): allowing degraded mounts [ 413.188814] BTRFS info (device dm-10): using free space tree [ 413.188815] BTRFS info (device dm-10): has skinny extents [ 413.189674] BTRFS warning (device dm-10): devid 5 uuid 068c6db3-3c30-4c97-b96b-5fe2d6c5d677 is missing [ 424.159486] BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1 [ 424.772640] BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 [ 434.749100] BTRFS info (device dm-10): found 6264 extents, stage: move data extents [ 477.703111] BTRFS info (device dm-10): found 6264 extents, stage: update data pointers [ 497.941482] BTRFS info (device dm-10): balance: ended with status: 0
The problematic block group was balanced successfully this time.
I balanced a few more successfully (without the -dconvert=single option), then decided to reboot under 5.10 just to see if I would hit this issue again. I didn't: the btrfs dev del worked correctly after the last 500G or so data was moved out of the drive.
This is the output of btrfs fi usage after I successfully balanced the problematic block group under the 5.6.17 kernel. Notice the multiple data profile, which is expected as I used the -dconvert balance option, and also the fact that apparently 3 chunks were allocated on new16T for this, even if only 1 seem to be used. We can tell because this is the first and only time the balance succeeded with the -dconvert option, hence these chunks are all under "data,single":
Overall:
Device size: 41.89TiB Device allocated: 21.74TiB Device unallocated: 20.14TiB Device missing: 9.09TiB Used: 21.71TiB Free (estimated): 10.08TiB (min: 10.07TiB) Data ratio: 2.00 Metadata ratio: 2.00 Global reserve: 512.00MiB (used: 0.00B) Multiple profiles: yes (data)
Data,single: Size:3.00GiB, Used:1.00GiB (33.34%)/dev/mapper/luks-new16T 3.00GiB
Data,RAID1: Size:10.83TiB, Used:10.83TiB (99.99%)/dev/mapper/luks-10Ta 7.14TiB/dev/mapper/luks-10Tb 7.10TiB
missing 482.00GiB/dev/mapper/luks-new16T 6.95TiB
Metadata,RAID1: Size:36.00GiB, Used:23.87GiB (66.31%)/dev/mapper/luks-10Tb 36.00GiB/dev/mapper/luks-ssd-mdata 36.00GiB
System,RAID1: Size:32.00MiB, Used:1.77MiB (5.52%)/dev/mapper/luks-10Ta 32.00MiB/dev/mapper/luks-10Tb 32.00MiB
Unallocated:
/dev/mapper/luks-10Ta 1.95TiB /dev/mapper/luks-10Tb 1.96TiB missing 8.62TiB
/dev/mapper/luks-ssd-mdata 11.29GiB/dev/mapper/luks-new16T 7.60TiB
I wasn't going to send an email to this ML because I knew I had nothing to reproduce the issue noww that it was "fixed", but now I think I'm bumping into the same issue on another FS, while rebalancing data after adding a drive, which happens to be the old 10T drive of the FS above.
The btrfs fi usage of this second FS is as follows:
Overall:
Device size: 25.50TiB Device allocated: 22.95TiB Device unallocated: 2.55TiB Device missing: 0.00B Used: 22.36TiB Free (estimated): 3.14TiB (min: 1.87TiB) Data ratio: 1.00 Metadata ratio: 2.00 Global reserve: 512.00MiB (used: 0.00B) Multiple profiles: no
Data,single: Size:22.89TiB, Used:22.29TiB (97.40%)/dev/mapper/luks-12T 10.91TiB
/dev/mapper/luks-3Ta 2.73TiB /dev/mapper/luks-3Tb 2.73TiB /dev/mapper/luks-10T 6.52TiB
Metadata,RAID1: Size:32.00GiB, Used:30.83GiB (96.34%) /dev/mapper/luks-ssd-mdata2 32.00GiB/dev/mapper/luks-10T 32.00GiB
System,RAID1: Size:32.00MiB, Used:2.44MiB (7.62%)/dev/mapper/luks-3Tb 32.00MiB/dev/mapper/luks-10T 32.00MiB
Unallocated:/dev/mapper/luks-12T 45.00MiB
/dev/mapper/luks-ssd-mdata2 4.00GiB /dev/mapper/luks-3Ta 1.02MiB /dev/mapper/luks-3Tb 2.97GiB /dev/mapper/luks-10T 2.54TiB
I can reproduce the problem reliably:
# btrfs bal start -dvrange=34625344765952..34625344765953 /tank ERROR: error during balancing '/tank': No such file or directory There may be more info in syslog - try dmesg | tail
[145979.563045] BTRFS info (device dm-10): balance: start -dvrange=34625344765952..34625344765953 [145979.585572] BTRFS info (device dm-10): relocating block group 34625344765952 flags data|raid1 [145990.396585] BTRFS info (device dm-10): found 167 extents, stage: move data extents [146002.236115] BTRFS info (device dm-10): balance: ended with status: -2
If anybody is interested in looking into this, this time I can leave the FS in this state. The issue is reproducible, and I can live without completing the balance for the next weeks or even months, as I don't think I'll need the currently unallocatable space soon.
I also made a btrfs-image of the FS, using btrfs-image -c 9 -t 4 -s -w. If it's of any use, I can drop it somewhere (51G).
I could try to bisect manually to find which version between 5.6.x and 5.10.1 started to behave like this, but on the first success, I won't know how to reproduce the issue a second time, as I'm not 100% sure it can be done solely with the btrfs-image.
Note that another user seem to have encoutered a similar issue in July with 5.8:
https://www.spinics.net/lists/linux-btrfs/msg103188.html
Regards,
Stéphane Lesimple.
Hi,
last week I had the same problem on a btrfs filesystem after updating to kernel 5.10.1. I have never had this problem before kernel 5.10.x. 5.9.x did now show any problem.
Dec 14 22:30:59 xxx kernel: BTRFS info (device sda2): scrub: started on devid 1 Dec 14 22:31:09 xxx kernel: BTRFS info (device sda2): scrub: finished on devid 1 with status: 0 Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: start -dusage=10 Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): relocating block group 71694286848 flags data Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): found 1058 extents, stage: move data extents Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: ended with status: -2
This is not a multidevice volume but a volume consisting of a single partition.
xxx ~ # btrfs fi df /u00
Data, single: total=10.01GiB, used=9.24GiB
System, single: total=4.00MiB, used=16.00KiB
Metadata, single: total=2.76GiB, used=1.10GiB
GlobalReserve, single: total=47.17MiB, used=0.00B
xxx ~ # btrfs device usage /u00/dev/sda2, ID: 1
Device size: 19.81GiB
Device slack: 0.00B
Data,single: 10.01GiB
Metadata,single: 2.76GiB
System,single: 4.00MiB
Unallocated: 7.04GiB
Quoted 169 lines · Stéphane Lesimple
Hello,
As part of the maintenance routine of one of my raid1 FS, a few days ago I was in the process of replacing a 10T drive with a 16T one.
So I first added the new 16T drive to the FS (btrfs dev add), then started a btrfs dev del.After a few days of balancing the block groups out of the old 10T drive, the balance aborted when around 500 GiB of data was still to be moved out of the drive:
Dec 21 14:18:40 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:18:54 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:19:16 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2Of course this also cancelled the device deletion, so after that the device was still part of the FS. I then tried to do a balance manually, in an attempt to reproduce the issue:
Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): balance: start -ddevid=5,limit=1 Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:28:29 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:28:46 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2There were of course still plenty of room on the FS, as I added a new 16T drive (a btrfs fi usage is further down this email), so it struck me as odd.
So, I tried to lower the reduncancy temporarily, expecting the balance of this block group to complete immediately given that there were already a copy of this data present on another drive:Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1 Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:39:00 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:39:17 nas kernel: BTRFS info (device dm-10): balance: ended with status: -2That didn't work.
I also tried to mount the FS in degraded mode, with the drive I wanted to remove missing, using btrfs dev del missing, but the balance still failed with the same error on the same block group.So, as I was running 5.10.1 just for a few days, I tried an older kernel: 5.6.17, and retried the balance once again (with still the drive voluntarily missing):
[ 413.188812] BTRFS info (device dm-10): allowing degraded mounts [ 413.188814] BTRFS info (device dm-10): using free space tree [ 413.188815] BTRFS info (device dm-10): has skinny extents [ 413.189674] BTRFS warning (device dm-10): devid 5 uuid 068c6db3-3c30-4c97-b96b-5fe2d6c5d677 is missing [ 424.159486] BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1 [ 424.772640] BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 [ 434.749100] BTRFS info (device dm-10): found 6264 extents, stage: move data extents [ 477.703111] BTRFS info (device dm-10): found 6264 extents, stage: update data pointers [ 497.941482] BTRFS info (device dm-10): balance: ended with status: 0The problematic block group was balanced successfully this time.
I balanced a few more successfully (without the -dconvert=single option), then decided to reboot under 5.10 just to see if I would hit this issue again. I didn't: the btrfs dev del worked correctly after the last 500G or so data was moved out of the drive.
This is the output of btrfs fi usage after I successfully balanced the problematic block group under the 5.6.17 kernel. Notice the multiple data profile, which is expected as I used the -dconvert balance option, and also the fact that apparently 3 chunks were allocated on new16T for this, even if only 1 seem to be used. We can tell because this is the first and only time the balance succeeded with the -dconvert option, hence these chunks are all under "data,single":
Overall:
Device size: 41.89TiB Device allocated: 21.74TiB Device unallocated: 20.14TiB Device missing: 9.09TiB Used: 21.71TiB Free (estimated): 10.08TiB (min: 10.07TiB) Data ratio: 2.00 Metadata ratio: 2.00 Global reserve: 512.00MiB (used: 0.00B) Multiple profiles: yes (data)Data,single: Size:3.00GiB, Used:1.00GiB (33.34%)
/dev/mapper/luks-new16T3.00GiBData,RAID1: Size:10.83TiB, Used:10.83TiB (99.99%)
/dev/mapper/luks-10Ta7.14TiB/dev/mapper/luks-10Tb7.10TiB
missing 482.00GiB/dev/mapper/luks-new16T6.95TiBMetadata,RAID1: Size:36.00GiB, Used:23.87GiB (66.31%)
/dev/mapper/luks-10Tb36.00GiB/dev/mapper/luks-ssd-mdata36.00GiBSystem,RAID1: Size:32.00MiB, Used:1.77MiB (5.52%)
/dev/mapper/luks-10Ta32.00MiB/dev/mapper/luks-10Tb32.00MiBUnallocated:
/dev/mapper/luks-10Ta 1.95TiB /dev/mapper/luks-10Tb 1.96TiB missing 8.62TiB
/dev/mapper/luks-ssd-mdata11.29GiB/dev/mapper/luks-new16T7.60TiBI wasn't going to send an email to this ML because I knew I had nothing to reproduce the issue noww that it was "fixed", but now I think I'm bumping into the same issue on another FS, while rebalancing data after adding a drive, which happens to be the old 10T drive of the FS above.
The btrfs fi usage of this second FS is as follows:
Overall:
Device size: 25.50TiB Device allocated: 22.95TiB Device unallocated: 2.55TiB Device missing: 0.00B Used: 22.36TiB Free (estimated): 3.14TiB (min: 1.87TiB) Data ratio: 1.00 Metadata ratio: 2.00 Global reserve: 512.00MiB (used: 0.00B) Multiple profiles: noData,single: Size:22.89TiB, Used:22.29TiB (97.40%)
/dev/mapper/luks-12T10.91TiB/dev/mapper/luks-3Ta 2.73TiB /dev/mapper/luks-3Tb 2.73TiB /dev/mapper/luks-10T 6.52TiBMetadata,RAID1: Size:32.00GiB, Used:30.83GiB (96.34%)
/dev/mapper/luks-ssd-mdata232.00GiB/dev/mapper/luks-10T32.00GiBSystem,RAID1: Size:32.00MiB, Used:2.44MiB (7.62%)
/dev/mapper/luks-3Tb32.00MiB/dev/mapper/luks-10T32.00MiBUnallocated:
/dev/mapper/luks-12T45.00MiB/dev/mapper/luks-ssd-mdata2 4.00GiB /dev/mapper/luks-3Ta 1.02MiB /dev/mapper/luks-3Tb 2.97GiB /dev/mapper/luks-10T 2.54TiBI can reproduce the problem reliably:
# btrfs bal start -dvrange=34625344765952..34625344765953 /tank ERROR: error during balancing '/tank': No such file or directory There may be more info in syslog - try dmesg | tail[145979.563045] BTRFS info (device dm-10): balance: start -dvrange=34625344765952..34625344765953 [145979.585572] BTRFS info (device dm-10): relocating block group 34625344765952 flags data|raid1 [145990.396585] BTRFS info (device dm-10): found 167 extents, stage: move data extents [146002.236115] BTRFS info (device dm-10): balance: ended with status: -2If anybody is interested in looking into this, this time I can leave the FS in this state. The issue is reproducible, and I can live without completing the balance for the next weeks or even months, as I don't think I'll need the currently unallocatable space soon.
I also made a btrfs-image of the FS, using btrfs-image -c 9 -t 4 -s -w. If it's of any use, I can drop it somewhere (51G).
I could try to bisect manually to find which version between 5.6.x and 5.10.1 started to behave like this, but on the first success, I won't know how to reproduce the issue a second time, as I'm not 100% sure it can be done solely with the btrfs-image.
Note that another user seem to have encoutered a similar issue in July with 5.8:
https://www.spinics.net/lists/linux-btrfs/msg103188.htmlRegards,
Stéphane Lesimple.
Quoted 28 lines · David Arendt
Hi,
last week I had the same problem on a btrfs filesystem after updating to kernel 5.10.1. I have never had this problem before kernel 5.10.x. 5.9.x did now show any problem.
Dec 14 22:30:59 xxx kernel: BTRFS info (device sda2): scrub: started on devid 1 Dec 14 22:31:09 xxx kernel: BTRFS info (device sda2): scrub: finished on devid 1 with status: 0 Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: start -dusage=10 Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): relocating block group 71694286848 flags data Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): found 1058 extents, stage: move data extents Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: endedwith status: -2
This is not a multidevice volume but a volume consisting of a single partition.
xxx ~ # btrfs fi df /u00
Data, single: total=10.01GiB, used=9.24GiB
System, single: total=4.00MiB, used=16.00KiB
Metadata, single: total=2.76GiB, used=1.10GiB
GlobalReserve, single: total=47.17MiB, used=0.00B
Quoted 8 lines
xxx ~ # btrfs device usage /u00
/dev/sda2, ID: 1
Device size: 19.81GiB
Device slack: 0.00B
Data,single: 10.01GiB
Metadata,single: 2.76GiB
System,single: 4.00MiB
Unallocated: 7.04GiB
This seems small enough, thus a btrfs-image dump would help.
Although there is a limit for btrfs-image dump, since it only contains metadata, when we try to balance data to reproduce the bug, it would easily cause data csum error and exit convert.
If possible, would you please try to take a dump with this branch?
https://github.com/adam900710/btrfs-progs/tree/image_data_dump
It provides a new option for btrfs-image, -d, which will also take the data.
Also, please keep in mind that, -d dump will contain data of your fs, thus if it contains confidential info, please use regular btrfs-image.
Thanks,
Qu
Quoted 212 lines · Stéphane Lesimple
Quoted 207 lines
Hello,
As part of the maintenance routine of one of my raid1 FS, a few days ago I was in the process
of replacing a 10T drive with a 16T one.
So I first added the new 16T drive to the FS (btrfs dev add), then started a btrfs dev del.After a few days of balancing the block groups out of the old 10T drive, the balance aborted when around 500 GiB of data was still to be moved out of the drive:
Dec 21 14:18:40 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:18:54 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:19:16 nas kernel: BTRFS info (device dm-10): balance: endedwith status: -2
Of course this also cancelled the device deletion, so after that the device was still part of the FS. I then tried to do a balance manually, in an attempt to reproduce the issue:
Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): balance: start -ddevid=5,limit=1 Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:28:29 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:28:46 nas kernel: BTRFS info (device dm-10): balance: endedwith status: -2
There were of course still plenty of room on the FS, as I added a new 16T drive
(a btrfs fi usage is further down this email), so it struck me as odd. So, I tried to lower the reduncancy temporarily, expecting the balance of this block group to
complete immediately given that there were already a copy of this data present on another drive:Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1 Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:39:00 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:39:17 nas kernel: BTRFS info (device dm-10): balance: endedwith status: -2
That didn't work.
I also tried to mount the FS in degraded mode, with the drive I wanted to remove missing,
using btrfs dev del missing, but the balance still failed with the same error on the same block group.So, as I was running 5.10.1 just for a few days, I tried an older
kernel: 5.6.17,
and retried the balance once again (with still the drive voluntarily missing):[ 413.188812] BTRFS info (device dm-10): allowing degraded mounts [ 413.188814] BTRFS info (device dm-10): using free space tree [ 413.188815] BTRFS info (device dm-10): has skinny extents [ 413.189674] BTRFS warning (device dm-10): devid 5 uuid 068c6db3-3c30-4c97-b96b-5fe2d6c5d677 is missing [ 424.159486] BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1 [ 424.772640] BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 [ 434.749100] BTRFS info (device dm-10): found 6264 extents, stage: move data extents [ 477.703111] BTRFS info (device dm-10): found 6264 extents, stage: update data pointers [ 497.941482] BTRFS info (device dm-10): balance: ended with status: 0The problematic block group was balanced successfully this time.
I balanced a few more successfully (without the -dconvert=single option), then decided to reboot under 5.10 just to see if I would hit this issue again.
I didn't: the btrfs dev del worked correctly after the last 500G or so data
was moved out of the drive.This is the output of btrfs fi usage after I successfully balanced the problematic block group under the 5.6.17 kernel. Notice the multiple data profile, which is expected as I used the -dconvert balance option, and also the fact that apparently 3 chunks were allocated on new16T for this, even if only 1 seem to be used. We can tell because this is the first and only time the balance succeeded with the -dconvert option, hence these chunks are all under "data,single":
Overall:
Device size: 41.89TiB
Device allocated: 21.74TiB
Device unallocated: 20.14TiB
Device missing: 9.09TiB
Used: 21.71TiB
Free (estimated): 10.08TiB (min: 10.07TiB)
Data ratio: 2.00
Metadata ratio: 2.00
Global reserve: 512.00MiB (used: 0.00B)
Multiple profiles: yes (data)Data,single: Size:3.00GiB, Used:1.00GiB (33.34%)
/dev/mapper/luks-new16T 3.00GiBData,RAID1: Size:10.83TiB, Used:10.83TiB (99.99%)
/dev/mapper/luks-10Ta 7.14TiB
/dev/mapper/luks-10Tb 7.10TiB
missing 482.00GiB
/dev/mapper/luks-new16T 6.95TiBMetadata,RAID1: Size:36.00GiB, Used:23.87GiB (66.31%) /dev/mapper/luks-10Tb 36.00GiB
/dev/mapper/luks-ssd-mdata36.00GiBSystem,RAID1: Size:32.00MiB, Used:1.77MiB (5.52%)
/dev/mapper/luks-10Ta 32.00MiB
/dev/mapper/luks-10Tb 32.00MiBUnallocated:
/dev/mapper/luks-10Ta 1.95TiB
/dev/mapper/luks-10Tb 1.96TiB
missing 8.62TiB/dev/mapper/luks-ssd-mdata11.29GiB
/dev/mapper/luks-new16T 7.60TiBI wasn't going to send an email to this ML because I knew I had nothing to reproduce the issue noww that it was "fixed", but now I think I'm bumping
into the same issue on another FS, while rebalancing data after adding a drive,
which happens to be the old 10T drive of the FS above.The btrfs fi usage of this second FS is as follows:
Overall:
Device size: 25.50TiB
Device allocated: 22.95TiB
Device unallocated: 2.55TiB
Device missing: 0.00B
Used: 22.36TiB
Free (estimated): 3.14TiB (min: 1.87TiB)
Data ratio: 1.00
Metadata ratio: 2.00
Global reserve: 512.00MiB (used: 0.00B)
Multiple profiles: noData,single: Size:22.89TiB, Used:22.29TiB (97.40%) /dev/mapper/luks-12T 10.91TiB /dev/mapper/luks-3Ta 2.73TiB /dev/mapper/luks-3Tb 2.73TiB /dev/mapper/luks-10T 6.52TiB
Metadata,RAID1: Size:32.00GiB, Used:30.83GiB (96.34%)
/dev/mapper/luks-ssd-mdata232.00GiB /dev/mapper/luks-10T 32.00GiBSystem,RAID1: Size:32.00MiB, Used:2.44MiB (7.62%)
/dev/mapper/luks-3Tb 32.00MiB
/dev/mapper/luks-10T 32.00MiBUnallocated:
/dev/mapper/luks-12T 45.00MiB
/dev/mapper/luks-ssd-mdata2 4.00GiB
/dev/mapper/luks-3Ta 1.02MiB
/dev/mapper/luks-3Tb 2.97GiB
/dev/mapper/luks-10T 2.54TiBI can reproduce the problem reliably:
# btrfs bal start -dvrange=34625344765952..34625344765953 /tank ERROR: error during balancing '/tank': No such file or directory There may be more info in syslog - try dmesg | tail[145979.563045] BTRFS info (device dm-10): balance: start -dvrange=34625344765952..34625344765953 [145979.585572] BTRFS info (device dm-10): relocating block group 34625344765952 flags data|raid1 [145990.396585] BTRFS info (device dm-10): found 167 extents, stage: move data extents [146002.236115] BTRFS info (device dm-10): balance: ended with status: -2If anybody is interested in looking into this, this time I can leave the FS in this state.
The issue is reproducible, and I can live without completing the balance for the next weeks
or even months, as I don't think I'll need the currently unallocatable space soon.I also made a btrfs-image of the FS, using btrfs-image -c 9 -t 4 -s -w. If it's of any use, I can drop it somewhere (51G).
I could try to bisect manually to find which version between 5.6.x and 5.10.1 started to behave
like this, but on the first success, I won't know how to reproduce the issue a second time, as
I'm not 100% sure it can be done solely with the btrfs-image.Note that another user seem to have encoutered a similar issue in July with 5.8:
https://www.spinics.net/lists/linux-btrfs/msg103188.htmlRegards,
Stéphane Lesimple.
Hi,
unfortunately the problem is no longer reproducible, probably due to writes happening in meantime. If you still want a btrfs-image, I can create one (unfortunately only without data as there is confidential data in it), but as the problem is currently no longer reproducible, I think it probably won't help.
Thanks in advance,
David Arendt
Quoted 268 lines · Qu Wenruo
Quoted 28 lines · David Arendt
Hi,
last week I had the same problem on a btrfs filesystem after updating to kernel 5.10.1. I have never had this problem before kernel 5.10.x. 5.9.x did now show any problem.
Dec 14 22:30:59 xxx kernel: BTRFS info (device sda2): scrub: started on devid 1 Dec 14 22:31:09 xxx kernel: BTRFS info (device sda2): scrub: finished on devid 1 with status: 0 Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: start -dusage=10 Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): relocating block group 71694286848 flags data Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): found 1058 extents, stage: move data extents Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: endedwith status: -2
This is not a multidevice volume but a volume consisting of a single partition.
xxx ~ # btrfs fi df /u00
Data, single: total=10.01GiB, used=9.24GiB
System, single: total=4.00MiB, used=16.00KiB
Metadata, single: total=2.76GiB, used=1.10GiB
GlobalReserve, single: total=47.17MiB, used=0.00BQuoted 8 lines
xxx ~ # btrfs device usage /u00
/dev/sda2, ID: 1
Device size: 19.81GiB
Device slack: 0.00B
Data,single: 10.01GiB
Metadata,single: 2.76GiB
System,single: 4.00MiB
Unallocated: 7.04GiBThis seems small enough, thus a btrfs-image dump would help.
Although there is a limit for btrfs-image dump, since it only contains metadata, when we try to balance data to reproduce the bug, it would easily cause data csum error and exit convert.
If possible, would you please try to take a dump with this branch?
https://github.com/adam900710/btrfs-progs/tree/image_data_dumpIt provides a new option for btrfs-image, -d, which will also take the data.
Also, please keep in mind that, -d dump will contain data of your fs, thus if it contains confidential info, please use regular btrfs-image.
Thanks,
QuQuoted 212 lines · Stéphane Lesimple
Quoted 207 lines
Hello,
As part of the maintenance routine of one of my raid1 FS, a few days ago I was in the process
of replacing a 10T drive with a 16T one.
So I first added the new 16T drive to the FS (btrfs dev add), then started a btrfs dev del.After a few days of balancing the block groups out of the old 10T drive, the balance aborted when around 500 GiB of data was still to be moved out of the drive:
Dec 21 14:18:40 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:18:54 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:19:16 nas kernel: BTRFS info (device dm-10): balance: endedwith status: -2
Of course this also cancelled the device deletion, so after that the device was still part of the FS. I then tried to do a balance manually, in an attempt to reproduce the issue:
Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): balance: start -ddevid=5,limit=1 Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:28:29 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:28:46 nas kernel: BTRFS info (device dm-10): balance: endedwith status: -2
There were of course still plenty of room on the FS, as I added a new 16T drive
(a btrfs fi usage is further down this email), so it struck me as odd. So, I tried to lower the reduncancy temporarily, expecting the balance of this block group to
complete immediately given that there were already a copy of this data present on another drive:Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1 Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:39:00 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:39:17 nas kernel: BTRFS info (device dm-10): balance: endedwith status: -2
That didn't work.
I also tried to mount the FS in degraded mode, with the drive I wanted to remove missing,
using btrfs dev del missing, but the balance still failed with the same error on the same block group.So, as I was running 5.10.1 just for a few days, I tried an older
kernel: 5.6.17,
and retried the balance once again (with still the drive voluntarily missing):[ 413.188812] BTRFS info (device dm-10): allowing degraded mounts [ 413.188814] BTRFS info (device dm-10): using free space tree [ 413.188815] BTRFS info (device dm-10): has skinny extents [ 413.189674] BTRFS warning (device dm-10): devid 5 uuid 068c6db3-3c30-4c97-b96b-5fe2d6c5d677 is missing [ 424.159486] BTRFS info (device dm-10): balance: start -dconvert=single,soft,devid=5,limit=1 [ 424.772640] BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 [ 434.749100] BTRFS info (device dm-10): found 6264 extents, stage: move data extents [ 477.703111] BTRFS info (device dm-10): found 6264 extents, stage: update data pointers [ 497.941482] BTRFS info (device dm-10): balance: ended with status: 0The problematic block group was balanced successfully this time.
I balanced a few more successfully (without the -dconvert=single option), then decided to reboot under 5.10 just to see if I would hit this issue again.
I didn't: the btrfs dev del worked correctly after the last 500G or so data
was moved out of the drive.This is the output of btrfs fi usage after I successfully balanced the problematic block group under the 5.6.17 kernel. Notice the multiple data profile, which is expected as I used the -dconvert balance option, and also the fact that apparently 3 chunks were allocated on new16T for this, even if only 1 seem to be used. We can tell because this is the first and only time the balance succeeded with the -dconvert option, hence these chunks are all under "data,single":
Overall:
Device size: 41.89TiB
Device allocated: 21.74TiB
Device unallocated: 20.14TiB
Device missing: 9.09TiB
Used: 21.71TiB
Free (estimated): 10.08TiB (min: 10.07TiB)
Data ratio: 2.00
Metadata ratio: 2.00
Global reserve: 512.00MiB (used: 0.00B)
Multiple profiles: yes (data)Data,single: Size:3.00GiB, Used:1.00GiB (33.34%)
/dev/mapper/luks-new16T 3.00GiBData,RAID1: Size:10.83TiB, Used:10.83TiB (99.99%)
/dev/mapper/luks-10Ta 7.14TiB
/dev/mapper/luks-10Tb 7.10TiB
missing 482.00GiB
/dev/mapper/luks-new16T 6.95TiBMetadata,RAID1: Size:36.00GiB, Used:23.87GiB (66.31%) /dev/mapper/luks-10Tb 36.00GiB
/dev/mapper/luks-ssd-mdata36.00GiBSystem,RAID1: Size:32.00MiB, Used:1.77MiB (5.52%)
/dev/mapper/luks-10Ta 32.00MiB
/dev/mapper/luks-10Tb 32.00MiBUnallocated:
/dev/mapper/luks-10Ta 1.95TiB
/dev/mapper/luks-10Tb 1.96TiB
missing 8.62TiB/dev/mapper/luks-ssd-mdata11.29GiB
/dev/mapper/luks-new16T 7.60TiBI wasn't going to send an email to this ML because I knew I had nothing to reproduce the issue noww that it was "fixed", but now I think I'm bumping
into the same issue on another FS, while rebalancing data after adding a drive,
which happens to be the old 10T drive of the FS above.The btrfs fi usage of this second FS is as follows:
Overall:
Device size: 25.50TiB
Device allocated: 22.95TiB
Device unallocated: 2.55TiB
Device missing: 0.00B
Used: 22.36TiB
Free (estimated): 3.14TiB (min: 1.87TiB)
Data ratio: 1.00
Metadata ratio: 2.00
Global reserve: 512.00MiB (used: 0.00B)
Multiple profiles: noData,single: Size:22.89TiB, Used:22.29TiB (97.40%) /dev/mapper/luks-12T 10.91TiB /dev/mapper/luks-3Ta 2.73TiB /dev/mapper/luks-3Tb 2.73TiB /dev/mapper/luks-10T 6.52TiB
Metadata,RAID1: Size:32.00GiB, Used:30.83GiB (96.34%)
/dev/mapper/luks-ssd-mdata232.00GiB /dev/mapper/luks-10T 32.00GiBSystem,RAID1: Size:32.00MiB, Used:2.44MiB (7.62%)
/dev/mapper/luks-3Tb 32.00MiB
/dev/mapper/luks-10T 32.00MiBUnallocated:
/dev/mapper/luks-12T 45.00MiB
/dev/mapper/luks-ssd-mdata2 4.00GiB
/dev/mapper/luks-3Ta 1.02MiB
/dev/mapper/luks-3Tb 2.97GiB
/dev/mapper/luks-10T 2.54TiBI can reproduce the problem reliably:
# btrfs bal start -dvrange=34625344765952..34625344765953 /tank ERROR: error during balancing '/tank': No such file or directory There may be more info in syslog - try dmesg | tail[145979.563045] BTRFS info (device dm-10): balance: start -dvrange=34625344765952..34625344765953 [145979.585572] BTRFS info (device dm-10): relocating block group 34625344765952 flags data|raid1 [145990.396585] BTRFS info (device dm-10): found 167 extents, stage: move data extents [146002.236115] BTRFS info (device dm-10): balance: ended with status: -2If anybody is interested in looking into this, this time I can leave the FS in this state.
The issue is reproducible, and I can live without completing the balance for the next weeks
or even months, as I don't think I'll need the currently unallocatable space soon.I also made a btrfs-image of the FS, using btrfs-image -c 9 -t 4 -s -w. If it's of any use, I can drop it somewhere (51G).
I could try to bisect manually to find which version between 5.6.x and 5.10.1 started to behave
like this, but on the first success, I won't know how to reproduce the issue a second time, as
I'm not 100% sure it can be done solely with the btrfs-image.Note that another user seem to have encoutered a similar issue in July with 5.8:
https://www.spinics.net/lists/linux-btrfs/msg103188.htmlRegards,
Stéphane Lesimple.
Quoted 3 lines · David Arendt
Hi,
unfortunately the problem is no longer reproducible, probably due to writes happening in meantime. If you still want a btrfs-image, I can create one (unfortunately only without data as there is confidential data in it), but as the problem is currently no longer reproducible, I think it probably won't help.
That's fine, at least you get your fs back to normal.
I tried several small balance locally, not reproduced, thus I guess it may be related to certain tree layout.
Anyway, I'll wait for another small enough and reproducible report.
Thanks,
Qu
Quoted 274 lines
Thanks in advance,
David ArendtQuoted 268 lines · Qu Wenruo
Quoted 28 lines · David Arendt
Hi,
last week I had the same problem on a btrfs filesystem after updating to kernel 5.10.1. I have never had this problem before kernel 5.10.x. 5.9.x did now show any problem.
Dec 14 22:30:59 xxx kernel: BTRFS info (device sda2): scrub: started on devid 1 Dec 14 22:31:09 xxx kernel: BTRFS info (device sda2): scrub: finished on devid 1 with status: 0 Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: start -dusage Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): relocating block group 71694286848 flags data Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): found 1058 extents, stage: move data extents Dec 14 22:33:16 xxx kernel: BTRFS info (device sda2): balance: endedwith status: -2
This is not a multidevice volume but a volume consisting of a single partition.
xxx ~ # btrfs fi df /u00
Data, single: total.01GiB, used=24GiB
System, single: total=00MiB, used.00KiB
Metadata, single: total=76GiB, used=10GiB
GlobalReserve, single: totalG.17MiB, used=00BQuoted 8 lines
xxx ~ # btrfs device usage /u00
/dev/sda2, ID: 1
Device size: 19.81GiB
Device slack: 0.00B
Data,single: 10.01GiB
Metadata,single: 2.76GiB
System,single: 4.00MiB
Unallocated: 7.04GiBThis seems small enough, thus a btrfs-image dump would help.
Although there is a limit for btrfs-image dump, since it only contains metadata, when we try to balance data to reproduce the bug, it would easily cause data csum error and exit convert.
If possible, would you please try to take a dump with this branch?
https://github.com/adam900710/btrfs-progs/tree/image_data_dumpIt provides a new option for btrfs-image, -d, which will also take the data.
Also, please keep in mind that, -d dump will contain data of your fs, thus if it contains confidential info, please use regular btrfs-image.
Thanks,
QuQuoted 212 lines · Stéphane Lesimple
Quoted 207 lines
Hello,
As part of the maintenance routine of one of my raid1 FS, a few days ago I was in the process
of replacing a 10T drive with a 16T one.
So I first added the new 16T drive to the FS (btrfs dev add), then started a btrfs dev del.After a few days of balancing the block groups out of the old 10T drive, the balance aborted when around 500 GiB of data was still to be moved out of the drive:
Dec 21 14:18:40 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:18:54 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:19:16 nas kernel: BTRFS info (device dm-10): balance: endedwith status: -2
Of course this also cancelled the device deletion, so after that the device was still part of the FS. I then tried to do a balance manually, in an attempt to reproduce the issue:
Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): balance: start -ddevid=limit= Dec 21 14:28:16 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:28:29 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:28:46 nas kernel: BTRFS info (device dm-10): balance: endedwith status: -2
There were of course still plenty of room on the FS, as I added a new 16T drive
(a btrfs fi usage is further down this email), so it struck me as odd. So, I tried to lower the reduncancy temporarily, expecting the balance of this block group to
complete immediately given that there were already a copy of this data present on another drive:Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): balance: start -dconvert=ngle,soft,devid=limit= Dec 21 14:38:50 nas kernel: BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 Dec 21 14:39:00 nas kernel: BTRFS info (device dm-10): found 6264 extents, stage: move data extents Dec 21 14:39:17 nas kernel: BTRFS info (device dm-10): balance: endedwith status: -2
That didn't work.
I also tried to mount the FS in degraded mode, with the drive I wanted to remove missing,
using btrfs dev del missing, but the balance still failed with the same error on the same block group.So, as I was running 5.10.1 just for a few days, I tried an older
kernel: 5.6.17,
and retried the balance once again (with still the drive voluntarily missing):[ 413.188812] BTRFS info (device dm-10): allowing degraded mounts [ 413.188814] BTRFS info (device dm-10): using free space tree [ 413.188815] BTRFS info (device dm-10): has skinny extents [ 413.189674] BTRFS warning (device dm-10): devid 5 uuid 068c6db3-3c30-4c97-b96b-5fe2d6c5d677 is missing [ 424.159486] BTRFS info (device dm-10): balance: start -dconvert=ngle,soft,devid=limit= [ 424.772640] BTRFS info (device dm-10): relocating block group 11115169841152 flags data|raid1 [ 434.749100] BTRFS info (device dm-10): found 6264 extents, stage: move data extents [ 477.703111] BTRFS info (device dm-10): found 6264 extents, stage: update data pointers [ 497.941482] BTRFS info (device dm-10): balance: ended with status: 0The problematic block group was balanced successfully this time.
I balanced a few more successfully (without the -dconvert=ngle option), then decided to reboot under 5.10 just to see if I would hit this issue again.
I didn't: the btrfs dev del worked correctly after the last 500G or so data
was moved out of the drive.This is the output of btrfs fi usage after I successfully balanced the problematic block group under the 5.6.17 kernel. Notice the multiple data profile, which is expected as I used the -dconvert balance option, and also the fact that apparently 3 chunks were allocated on new16T for this, even if only 1 seem to be used. We can tell because this is the first and only time the balance succeeded with the -dconvert option, hence these chunks are all under "data,single":
Overall:
Device size: 41.89TiB
Device allocated: 21.74TiB
Device unallocated: 20.14TiB
Device missing: 9.09TiB
Used: 21.71TiB
Free (estimated): 10.08TiB (min: 10.07TiB)
Data ratio: 2.00
Metadata ratio: 2.00
Global reserve: 512.00MiB (used: 0.00B)
Multiple profiles: yes (data)Data,single: Size:3.00GiB, Used:1.00GiB (33.34%)
/dev/mapper/luks-new16T 3.00GiBData,RAID1: Size:10.83TiB, Used:10.83TiB (99.99%)
/dev/mapper/luks-10Ta 7.14TiB
/dev/mapper/luks-10Tb 7.10TiB
missing 482.00GiB
/dev/mapper/luks-new16T 6.95TiBMetadata,RAID1: Size:36.00GiB, Used:23.87GiB (66.31%) /dev/mapper/luks-10Tb 36.00GiB
/dev/mapper/luks-ssd-mdata36.00GiBSystem,RAID1: Size:32.00MiB, Used:1.77MiB (5.52%)
/dev/mapper/luks-10Ta 32.00MiB
/dev/mapper/luks-10Tb 32.00MiBUnallocated:
/dev/mapper/luks-10Ta 1.95TiB
/dev/mapper/luks-10Tb 1.96TiB
missing 8.62TiB/dev/mapper/luks-ssd-mdata11.29GiB
/dev/mapper/luks-new16T 7.60TiBI wasn't going to send an email to this ML because I knew I had nothing to reproduce the issue noww that it was "fixed", but now I think I'm bumping
into the same issue on another FS, while rebalancing data after adding a drive,
which happens to be the old 10T drive of the FS above.The btrfs fi usage of this second FS is as follows:
Overall:
Device size: 25.50TiB
Device allocated: 22.95TiB
Device unallocated: 2.55TiB
Device missing: 0.00B
Used: 22.36TiB
Free (estimated): 3.14TiB (min: 1.87TiB)
Data ratio: 1.00
Metadata ratio: 2.00
Global reserve: 512.00MiB (used: 0.00B)
Multiple profiles: noData,single: Size:22.89TiB, Used:22.29TiB (97.40%) /dev/mapper/luks-12T 10.91TiB /dev/mapper/luks-3Ta 2.73TiB /dev/mapper/luks-3Tb 2.73TiB /dev/mapper/luks-10T 6.52TiB
Metadata,RAID1: Size:32.00GiB, Used:30.83GiB (96.34%)
/dev/mapper/luks-ssd-mdata232.00GiB /dev/mapper/luks-10T 32.00GiBSystem,RAID1: Size:32.00MiB, Used:2.44MiB (7.62%)
/dev/mapper/luks-3Tb 32.00MiB
/dev/mapper/luks-10T 32.00MiBUnallocated:
/dev/mapper/luks-12T 45.00MiB
/dev/mapper/luks-ssd-mdata2 4.00GiB
/dev/mapper/luks-3Ta 1.02MiB
/dev/mapper/luks-3Tb 2.97GiB
/dev/mapper/luks-10T 2.54TiBI can reproduce the problem reliably:
# btrfs bal start -dvrange4625344765952..34625344765953 /tank ERROR: error during balancing '/tank': No such file or directory There may be more info in syslog - try dmesg | tail[145979.563045] BTRFS info (device dm-10): balance: start -dvrange4625344765952..34625344765953 [145979.585572] BTRFS info (device dm-10): relocating block group 34625344765952 flags data|raid1 [145990.396585] BTRFS info (device dm-10): found 167 extents, stage: move data extents [146002.236115] BTRFS info (device dm-10): balance: ended with status: -2If anybody is interested in looking into this, this time I can leave the FS in this state.
The issue is reproducible, and I can live without completing the balance for the next weeks
or even months, as I don't think I'll need the currently unallocatable space soon.I also made a btrfs-image of the FS, using btrfs-image -c 9 -t 4 -s -w. If it's of any use, I can drop it somewhere (51G).
I could try to bisect manually to find which version between 5.6.x and 5.10.1 started to behave
like this, but on the first success, I won't know how to reproduce the issue a second time, as
I'm not 100% sure it can be done solely with the btrfs-image.Note that another user seem to have encoutered a similar issue in July with 5.8:
https://www.spinics.net/lists/linux-btrfs/msg103188.htmlRegards,
Stéphane Lesimple.
Quoted 6 lines
Quoted 5 lines
unfortunately the problem is no longer reproducible, probably due to writes happening in meantime. If you still want a btrfs-image, I can create one (unfortunately only without data as there is confidential data in it), but as the problem is currently no longer reproducible, I think it probably won't help.
Quoted 6 lines
That's fine, at least you get your fs back to normal.
I tried several small balance locally, not reproduced, thus I guess it may be related to certain tree layout.
Anyway, I'll wait for another small enough and reproducible report.
This is still reproducible on my FS, and I have the btrfs-image. I can easily upload it somewhere, but of course I understand downloading an image of 51G can be impractical.
An other way might be: as I know which block group is causing the problem, as per the dmesg, maybe I can dump only the part of the metadata relevant to this block group?
In any case I can run commands on this system, compile a custom btrfs-progs or a custom kernel with whatever you want me to try, and reboot as many times as necessary (this is not a production server).
I know it fails in relocate_block_group(), which returns -2, I'm currently adding a couple printk's here and there to try to pinpoint that better.
Regards,
Stéphane.
Quoted 2 lines
I know it fails in
relocate_block_group(), which returns -2, I'm currently adding a couple printk's here and there to try to pinpoint that better.
Okay, so btrfs_relocate_block_group() starts with stage MOVE_DATA_EXTENTS, which completes successfully, as relocate_block_group() returns 0:
BTRFS info (device <unknown>): relocate_block_group: prepare_to_realocate = 0
BTRFS info (device <unknown>): relocate_block_group loop: progress = 1, btrfs_start_transaction = ok [...]
BTRFS info (device <unknown>): relocate_block_group loop: progress = 168, btrfs_start_transaction = ok BTRFS info (device <unknown>): relocate_block_group: returning err = 0
BTRFS info (device dm-10): stage = move data extents, relocate_block_group = 0
BTRFS info (device dm-10): found 167 extents, stage: move data extents
Then it proceeds to the UPDATE_DATA_PTRS stage and calls relocate_block_group() again. This time it'll fail at the 92th iteration of the loop:
BTRFS info (device <unknown>): relocate_block_group loop: progress = 92, btrfs_start_transaction = ok
BTRFS info (device <unknown>): relocate_block_group loop: extents_found = 92, item_size(53) >= sizeof(*ei)(24), flags = 1, ret = 0 BTRFS info (device <unknown>): add_data_references: btrfs_find_all_leafs = 0
BTRFS info (device <unknown>): add_data_references loop: read_tree_block ok
BTRFS info (device <unknown>): add_data_references loop: delete_v1_space_cache = -2
BTRFS info (device <unknown>): relocate_block_group loop: add_data_references = -2
Then the -ENOENT goes all the way up the call stack and aborts the balance.
So it fails in delete_v1_space_cache(), though it is worth noting that the FS we're talking about is actually using space_cache v2.
Does it help? Shall I dig deeper?
Regards,
Stéphane.
Quoted 14 lines · Stéphane Lesimple
Quoted 2 lines
I know it fails in
relocate_block_group(), which returns -2, I'm currently adding a couple printk's here and there to try to pinpoint that better.Okay, so
btrfs_relocate_block_group()starts with stageMOVE_DATA_EXTENTS, which completes successfully, asrelocate_block_group()returns 0:BTRFS info (device <unknown>):
relocate_block_group:prepare_to_realocate= 0
BTRFS info (device <unknown>):relocate_block_grouploop: progress = 1,btrfs_start_transaction= ok [...]
BTRFS info (device <unknown>):relocate_block_grouploop: progress = 168,btrfs_start_transaction= ok BTRFS info (device <unknown>):relocate_block_group: returning err = 0
BTRFS info (device dm-10): stage = move data extents,relocate_block_group= 0
BTRFS info (device dm-10): found 167 extents, stage: move data extents
Quoted 8 lines
Then it proceeds to the
UPDATE_DATA_PTRSstage and callsrelocate_block_group()again. This time it'll fail at the 92th iteration of the loop:BTRFS info (device <unknown>):
relocate_block_grouploop: progress = 92,btrfs_start_transaction= ok
BTRFS info (device <unknown>):relocate_block_grouploop:extents_found= 92,item_size(53)>= sizeof(*ei)(24), flags = 1, ret = 0 BTRFS info (device <unknown>):add_data_references:btrfs_find_all_leafs= 0
BTRFS info (device <unknown>):add_data_referencesloop:read_tree_blockok
BTRFS info (device <unknown>):add_data_referencesloop:delete_v1_space_cache= -2
Damn it, if we find no v1 space cache for the block group, it means we're fine to continue...
Quoted 6 lines
BTRFS info (device <unknown>):
relocate_block_grouploop:add_data_references= -2Then the
-ENOENTgoes all the way up the call stack and aborts the balance.So it fails in
delete_v1_space_cache(), though it is worth noting that the FS we're talking about is actually usingspace_cachev2.
Space cache v2, no wonder no v1 space cache.
Quoted 2 lines
Does it help? Shall I dig deeper?
You're already at the point!
Mind me to craft a fix with your signed-off-by?
Thanks,
Qu
Quoted 5 lines
Regards,
Stéphane.
Quoted 51 lines · Qu Wenruo
Quoted 24 lines · Stéphane Lesimple
Quoted 3 lines
I know it fails in
relocate_block_group(), which returns -2, I'm currently
adding a couple printk's here and there to try to pinpoint that better.Okay, so
btrfs_relocate_block_group()starts with stageMOVE_DATA_EXTENTS, which
completes successfully, asrelocate_block_group()returns 0:BTRFS info (device <unknown>):
relocate_block_group:prepare_to_realocate= 0
BTRFS info (device <unknown>):relocate_block_grouploop: progress = 1,btrfs_start_transaction= ok
[...]
BTRFS info (device <unknown>):relocate_block_grouploop: progress = 168,btrfs_start_transaction= ok
BTRFS info (device <unknown>):relocate_block_group: returning err = 0 BTRFS info (device dm-10): stage = move data extents,relocate_block_group= 0
BTRFS info (device dm-10): found 167 extents, stage: move data extentsThen it proceeds to the
UPDATE_DATA_PTRSstage and callsrelocate_block_group()
again. This time it'll fail at the 92th iteration of the loop:Quoted 10 lines
BTRFS info (device <unknown>):
relocate_block_grouploop: progress = 92,btrfs_start_transaction= ok
BTRFS info (device <unknown>):relocate_block_grouploop:extents_found= 92,item_size(53)>= sizeof(*ei)(24), flags = 1, ret = 0 BTRFS info (device <unknown>):add_data_references:btrfs_find_all_leafs= 0
BTRFS info (device <unknown>):add_data_referencesloop:read_tree_blockok
BTRFS info (device <unknown>):add_data_referencesloop:delete_v1_space_cache= -2Damn it, if we find no v1 space cache for the block group, it means we're fine to continue...
Quoted 9 lines
BTRFS info (device <unknown>):
relocate_block_grouploop:add_data_references= -2Then the
-ENOENTgoes all the way up the call stack and aborts the balance.So it fails in
delete_v1_space_cache(), though it is worth noting that the
FS we're talking about is actually usingspace_cachev2.
Quoted 8 lines
Space cache v2, no wonder no v1 space cache.
Quoted 2 lines
Does it help? Shall I dig deeper?
You're already at the point!
Mind me to craft a fix with your signed-off-by?
The problem is more complex than I thought, but still we at least have some workaround.
Firstly, this happens when an old fs get v2 space cache enabled, but still has v1 space cache left.
Newer v2 mount should cleanup v1 properly, but older kernel doesn't do the proper cleaning, thus left some v1 cache.
Then we call btrfs balance on such old fs, leading to the -ENOENT error. We can't ignore the error, as we have no way to relocate such left over v1 cache (normally we delete it completely, but with v2 cache, we can't).
So what I can do is only to add a warning message to the problem.
To solve your problem, I also submitted a patch to btrfs-progs, to force v1 space cache cleaning even if the fs has v2 space cache enabled.
Or, you can disable v2 space cache first, using "btrfs check --clear-space-cache v2" first, then "btrfs check --clear-space_cache v1", and finally mount the fs with "space_cache=v2" again.
To verify there is no space cache v1 left, you can run the following command to verify:
# btrfs ins dump-tree -t root <device> | grep EXTENT_DATA
It should output nothing.
Then please try if you can balance all your data.
Thanks,
Qu
Quoted 9 lines
Thanks,
QuQuoted 5 lines
Regards,
Stéphane.
Hi,
Just for information: On my system the error appeared on a filesystem using space cache v1. I think my problem might then be unrelated to this one. If it will happen again, I will try to collect more information. Maybe a should try a clear_cache to ensure that the space cache is not wrong.
Bye,
David Arendt
Quoted 106 lines · Qu Wenruo
Quoted 51 lines · Qu Wenruo
Quoted 24 lines · Stéphane Lesimple
Quoted 3 lines
I know it fails in
relocate_block_group(), which returns -2, I'm currently
adding a couple printk's here and there to try to pinpoint that better.Okay, so
btrfs_relocate_block_group()starts with stageMOVE_DATA_EXTENTS, which
completes successfully, asrelocate_block_group()returns 0:BTRFS info (device <unknown>):
relocate_block_group:prepare_to_realocate= 0
BTRFS info (device <unknown>):relocate_block_grouploop: progress = 1,btrfs_start_transaction= ok
[...]
BTRFS info (device <unknown>):relocate_block_grouploop: progress = 168,btrfs_start_transaction= ok
BTRFS info (device <unknown>):relocate_block_group: returning err = 0 BTRFS info (device dm-10): stage = move data extents,relocate_block_group= 0
BTRFS info (device dm-10): found 167 extents, stage: move data extentsThen it proceeds to the
UPDATE_DATA_PTRSstage and callsrelocate_block_group()
again. This time it'll fail at the 92th iteration of the loop:Quoted 10 lines
BTRFS info (device <unknown>):
relocate_block_grouploop: progress = 92,btrfs_start_transaction= ok
BTRFS info (device <unknown>):relocate_block_grouploop:extents_found= 92,item_size(53)>= sizeof(*ei)(24), flags = 1, ret = 0 BTRFS info (device <unknown>):add_data_references:btrfs_find_all_leafs= 0
BTRFS info (device <unknown>):add_data_referencesloop:read_tree_blockok
BTRFS info (device <unknown>):add_data_referencesloop:delete_v1_space_cache= -2Damn it, if we find no v1 space cache for the block group, it means we're fine to continue...
Quoted 9 lines
BTRFS info (device <unknown>):
relocate_block_grouploop:add_data_references= -2Then the
-ENOENTgoes all the way up the call stack and aborts the balance.So it fails in
delete_v1_space_cache(), though it is worth noting that the
FS we're talking about is actually usingspace_cachev2.Quoted 8 lines
Space cache v2, no wonder no v1 space cache.
Quoted 2 lines
Does it help? Shall I dig deeper?
You're already at the point!
Mind me to craft a fix with your signed-off-by?
The problem is more complex than I thought, but still we at least have some workaround.
Firstly, this happens when an old fs get v2 space cache enabled, but still has v1 space cache left.
Newer v2 mount should cleanup v1 properly, but older kernel doesn't do the proper cleaning, thus left some v1 cache.
Then we call btrfs balance on such old fs, leading to the
-ENOENTerror. We can't ignore the error, as we have no way to relocate such left over v1 cache (normally we delete it completely, but with v2 cache, we can't).So what I can do is only to add a warning message to the problem.
To solve your problem, I also submitted a patch to btrfs-progs, to force v1 space cache cleaning even if the fs has v2 space cache enabled.
Or, you can disable v2 space cache first, using "btrfs check --clear-space-cache v2" first, then "btrfs check --clear-space_cache v1", and finally mount the fs with "space_cache=v2" again.
To verify there is no space cache v1 left, you can run the following command to verify:
# btrfs ins dump-tree -t root <device> | grep EXTENT_DATAIt should output nothing.
Then please try if you can balance all your data.
Thanks,
QuQuoted 9 lines
Thanks,
QuQuoted 5 lines
Regards,
Stéphane.
Quoted 3 lines · David Arendt
Hi,
Just for information: On my system the error appeared on a filesystem using space cache v1. I think my problem might then be unrelated to this one.
Then this is more interesting.
There are two locations which can return -ENOENT in delete_v1_space_cache():
- No file extent found
This means something is wrong in backref walk.
I don't believe it's even possible, or qgroup and balance will be completely broken.
- delete_block_group_cache() failed to grab the free space cache inode There is another possibility that, we have free space cache inode in commit root (where data relocation reads from), but it's not in our current root.
In that case, -ENOENT is safe to ignore.
I guess you may hit the 2nd case, as your next balance finishes without problem.
Quoted 1 line
If it will happen again, I will try to collect more information. Maybe a should try a
clear_cacheto ensure that the space cache is not wrong.
Clear_cache itself won't remove all the existing cache, it just remove the caches when the block group gets dirty.
Thus we use btrfs-check to remove free space cache completely.
Thanks,
Qu
Quoted 112 lines
Bye,
David ArendtQuoted 106 lines · Qu Wenruo
Quoted 51 lines · Qu Wenruo
Quoted 24 lines · Stéphane Lesimple
Quoted 3 lines
I know it fails in
relocate_block_group(), which returns -2, I'm currently
adding a couple printk's here and there to try to pinpoint that better.Okay, so
btrfs_relocate_block_group()starts with stageMOVE_DATA_EXTENTS, which
completes successfully, asrelocate_block_group()returns 0:BTRFS info (device <unknown>):
relocate_block_group:prepare_to_realocate= 0
BTRFS info (device <unknown>):relocate_block_grouploop: progress = 1,btrfs_start_transaction= ok
[...]
BTRFS info (device <unknown>):relocate_block_grouploop: progress = 168,btrfs_start_transaction= ok
BTRFS info (device <unknown>):relocate_block_group: returning err = 0 BTRFS info (device dm-10): stage = move data extents,relocate_block_group= 0
BTRFS info (device dm-10): found 167 extents, stage: move data extentsThen it proceeds to the
UPDATE_DATA_PTRSstage and callsrelocate_block_group()
again. This time it'll fail at the 92th iteration of the loop:Quoted 10 lines
BTRFS info (device <unknown>):
relocate_block_grouploop: progress = 92,btrfs_start_transaction= ok
BTRFS info (device <unknown>):relocate_block_grouploop:extents_found= 92,item_size(53)>= sizeof(*ei)(24), flags = 1, ret = 0 BTRFS info (device <unknown>):add_data_references:btrfs_find_all_leafs= 0
BTRFS info (device <unknown>):add_data_referencesloop:read_tree_blockok
BTRFS info (device <unknown>):add_data_referencesloop:delete_v1_space_cache= -2Damn it, if we find no v1 space cache for the block group, it means we're fine to continue...
Quoted 9 lines
BTRFS info (device <unknown>):
relocate_block_grouploop:add_data_references= -2Then the
-ENOENTgoes all the way up the call stack and aborts the balance.So it fails in
delete_v1_space_cache(), though it is worth noting that the
FS we're talking about is actually usingspace_cachev2.Quoted 8 lines
Space cache v2, no wonder no v1 space cache.
Quoted 2 lines
Does it help? Shall I dig deeper?
You're already at the point!
Mind me to craft a fix with your signed-off-by?
The problem is more complex than I thought, but still we at least have some workaround.
Firstly, this happens when an old fs get v2 space cache enabled, but still has v1 space cache left.
Newer v2 mount should cleanup v1 properly, but older kernel doesn't do the proper cleaning, thus left some v1 cache.
Then we call btrfs balance on such old fs, leading to the
-ENOENTerror. We can't ignore the error, as we have no way to relocate such left over v1 cache (normally we delete it completely, but with v2 cache, we can't).So what I can do is only to add a warning message to the problem.
To solve your problem, I also submitted a patch to btrfs-progs, to force v1 space cache cleaning even if the fs has v2 space cache enabled.
Or, you can disable v2 space cache first, using "btrfs check --clear-space-cache v2" first, then "btrfs check --clear-space_cache v1", and finally mount the fs with "space_cache=v2" again.
To verify there is no space cache v1 left, you can run the following command to verify:
# btrfs ins dump-tree -t root <device> | grep EXTENT_DATAIt should output nothing.
Then please try if you can balance all your data.
Thanks,
QuQuoted 9 lines
Thanks,
QuQuoted 5 lines
Regards,
Stéphane.
Quoted 1 line
Quoted 1 line
Mind me to craft a fix with your signed-off-by?
Sure!
Quoted 22 lines
The problem is more complex than I thought, but still we at least have some workaround.
Firstly, this happens when an old fs get v2 space cache enabled, but still has v1 space cache left.
Newer v2 mount should cleanup v1 properly, but older kernel doesn't do the proper cleaning, thus left some v1 cache.
Then we call btrfs balance on such old fs, leading to the
-ENOENTerror. We can't ignore the error, as we have no way to relocate such left over v1 cache (normally we delete it completely, but with v2 cache, we can't).So what I can do is only to add a warning message to the problem.
To solve your problem, I also submitted a patch to btrfs-progs, to force v1 space cache cleaning even if the fs has v2 space cache enabled.
Or, you can disable v2 space cache first, using "btrfs check --clear-space-cache v2" first, then "btrfs check --clear-space_cache v1", and finally mount the fs with "space_cache=v2" again.
Quoted 8 lines
To verify there is no space cache v1 left, you can run the following command to verify:
# btrfs ins dump-tree -t root <device> | grep EXTENT_DATAIt should output nothing.
Then please try if you can balance all your data.
Your analysis is correct, I do have v1 leftovers as I commented on the [PATCH] you've sent.
Now, fixing the FS:
# btrfs check --clear-space-cache v2 /dev/mapper/luks-tank-mdata Opening filesystem to check... Checking filesystem on /dev/mapper/luks-tank-mdata UUID: 428b20da-dcb1-403e-b407-ba984fd07ebd Clear free space cache v2 Segmentation fault
Wow, okay. That's unexpected.
# btrfs --version btrfs-progs v5.9
(gdb) r
Starting program: /usr/local/bin/btrfs check --clear-space-cache v2 /dev/mapper/luks-tank-mdata [Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1". Opening filesystem to check...
Checking filesystem on /dev/mapper/luks-tank-mdata
UUID: 428b20da-dcb1-403e-b407-ba984fd07ebd
Clear free space cache v2
Program received signal SIGSEGV, Segmentation fault.balance_level (level=<optimized out>, path=0x555555649490, root=0x555555645da0, trans=<optimized out>) at kernel-shared/ctree.c:930
930 root_sub_used(root, right->len);
(gdb) bt
#0 balance_level (level=<optimized out>, path=0x555555649490, root=0x555555645da0, trans=<optimized out>) at kernel-shared/ctree.c:930
#1 btrfs_search_slot (trans=trans@entry=0x55555e8b4d30, root=root@entry=0x555555645da0, key=key@entry=0x7fffffffe000, p=p@entry=0x555555649490, ins_len=ins_len@entry=-1, cow=cow@entry=1)
at kernel-shared/ctree.c:1320 #2 0x00005555555e3da7 in clear_free_space_tree (root=0x555555645da0, trans=0x55555e8b4d30) at kernel-shared/free-space-tree.c:1161 #3 btrfs_clear_free_space_tree (fs_info=<optimized out>) at kernel-shared/free-space-tree.c:1201 #4 0x000055555558cd5f in do_clear_free_space_cache (clear_version=clear_version@entry=2) at check/main.c:9872 #5 0x000055555559acce in cmd_check (cmd=0x555555638900 <cmd_struct_check>, argc=<optimized out>, argv=0x7fffffffe490) at check/main.c:10194 #6 0x000055555556ae88 in cmd_execute (argv=0x7fffffffe490, argc=4, cmd=0x555555638900 <cmd_struct_check>) at cmds/commands.h:125 #7 main (argc=4, argv=0x7fffffffe490) at btrfs.c:402
(gdb)
Can v1 leftovers provoke this?
The patch you've sent for btrfs-progs might fix my problem as I wouldn't need to remove space_cache v2 first, so I may not hit this bug, but if you're interested in looking into this one too, we might kill one bird with two stones!
I'm leaving my FS as is waiting for your reply,
Regards,
Stéphane.
Qu Wenruo - 29.12.20, 01:44:07 CET:
Quoted 10 lines
So what I can do is only to add a warning message to the problem.
To solve your problem, I also submitted a patch to btrfs-progs, to force v1 space cache cleaning even if the fs has v2 space cache enabled.
Or, you can disable v2 space cache first, using "btrfs check --clear-space-cache v2" first, then "btrfs check --clear-space_cache v1", and finally mount the fs with "space_cache=v2" again.
Quoted 6 lines
To verify there is no space cache v1 left, you can run the following command to verify:
# btrfs ins dump-tree -t root <device> | grep EXTENT_DATAIt should output nothing.
I have v1 space_cache stuff on filesystems which use v2 space_cache as well, so…
the fully working way to completely switch to spacecache_v2 for any BTRFS filesystem with space cache v1, is what you wrote above?
Or would it be more straight forward than that with a newer kernel?
Best,
Signature · 1 line
Martin